Game Theory and DQN-Based Network Defense Decision-Making for Terminal Security in Power Distribution System

The increase of terminal devices in the automated power distribution system expands the attack surface for malicious adversaries, which can severely impair the reliability and stability of the power grids. As cyber-attacks are empowered by diverse and elaborate techniques, relying solely on traditional passive defense methods is insufficient to address the evolving security threats. It is imperative to develop adaptive and active approaches to defend against the cyber-attacks aimed at terminal devices in the power distribution system. Motivated by this idea, we focus on the network defense decision-making approach. Effective defense strategies not only mitigate the risk of disruptive cyber-attacks, such as data breaches and system compromises, but also ensure the continuous operation of essential services in the power system. Though researchers have made efforts to explore game-theoretic defense-decision methods for power systems, there are still shortcoming in defining rewards and state transitions. In this paper, we utilize the stochastic game to model the confrontation between the attacker and defender. We design a hybrid reward function to describe the rewards in the two-player game from various perspectives. We also introduce a deep Q-network to learn the high-dimensional state transitions and achieve the Nash equilibrium. Experimental results based on the simulated network are presented to demonstrate the effectiveness of the proposed approach.